Sandwich structure plate core debonding mechanical property intelligent prediction method

Abstract

This invention provides an intelligent prediction method for the mechanical properties of core debonding in sandwich structures, comprising: Step 1: constructing a sandwich structure simulation model library through finite element parametric modeling; Step 2: constructing a core debonding mechanical property database based on the sandwich structure simulation model library; Step 3: using a deep learning model to intelligently predict the core debonding mechanical properties based on the core debonding mechanical property database.

Description

Technical Field

[0001] This invention relates to the field of mechanical property prediction, and more specifically to a data-driven method for predicting the mechanical properties of core debonding in sandwich structure panels. Background Technology

[0002] Sandwich structures are a core load-bearing structural form for modern lightweight high-end equipment. A typical configuration consists of two layers of high-strength fiber-reinforced resin-based composite material panels bonded together with a lightweight porous core material in the middle using structural adhesive. The panels primarily bear in-plane tensile, compressive, and bending stresses, while the lightweight core material mainly bears lateral shear loads and provides stable support. The overall structure possesses comprehensive advantages unmatched by traditional metal structures, including high specific strength, high specific stiffness, excellent fatigue resistance, vibration reduction, noise reduction, heat insulation, and flame retardancy. In recent years, with the continuous upgrading of development demands for lightweight aerospace vehicles, ultra-long aspect ratio design of large wind turbine blades, energy conservation and weight reduction in rail transit vehicles, and corrosion protection and lightweighting of ship superstructures, the application coverage of sandwich structures has been increasing year by year, gradually replacing traditional aluminum alloy and steel structures as the preferred material system for main load-bearing components.

[0003] In actual engineering service, the weakest point of a sandwich structure is not the panel and the core material itself, but the adhesive interface layer between the panel and the core material. The adhesive interface between the panel and the core material is highly susceptible to delamination defects under the coupled effects of multiple factors: during the manufacturing stage, process defects such as insufficient adhesive wetting, uneven curing temperature and pressure, air bubbles, and impurities can pre-embed initial micro-delamination defects; during the assembly and use stage, secondary processing such as structural drilling, riveting, and grinding can cause the propagation of micro-cracks at the interface; during long-term service, complex working conditions such as high and low temperature cycling, damp heat aging corrosion, ultraviolet radiation, rainwater infiltration, alternating fatigue loads, low-speed impact, and vibration coupling can continuously promote the expansion of micro-delamination defects into macroscopic large-area delamination.

[0004] Once debonding defects form, they directly sever the collaborative force transmission mechanism between the panel and the core material, completely altering the force transmission path of the sandwich structure. A complete sandwich structure follows the classical mechanical principle of "panel bearing bending, core bearing shear" under stress, resulting in uniform overall structural stiffness and continuous stress distribution. However, after debonding defects occur, the panel in the debonded area loses the support of the core material, resulting in a locally suspended stress state. This leads to a significant decrease in bending stiffness, significant local stress concentration, and a sharp drop in the buckling critical load, making it highly susceptible to severe failures such as local collapse, panel wrinkling, structural buckling instability, and overall fracture failure. For large, critical load-bearing structures such as aircraft and large wind turbine blades, progressive structural failure caused by localized panel-core debonding is a core cause of unexpected equipment failures, increased maintenance costs, and shortened service life. In severe cases, it can lead to flight safety accidents and wind turbine collapses, posing extremely high engineering risks. Therefore, accurately characterizing the evolution of mechanical properties of sandwich structures under different debonding dimensions, core thicknesses, layup angles, and service conditions, and quickly obtaining key mechanical parameters such as ultimate bearing capacity, buckling stability, stiffness degradation rate, and fatigue life, is the core technological foundation for achieving structural damage classification, remaining service life assessment, preventive maintenance, and structural optimization design.

[0005] Currently, the academic and industrial communities both domestically and internationally have developed two main mature technical solutions for evaluating and predicting the debonding mechanical properties of sandwich structure core panels: full-size solid structure testing and refined finite element numerical calculation. Mechanical testing is the traditional benchmark testing method in the industry. Based on actual engineering structural parameters, standard composite sandwich specimens with different panel thicknesses, core material thicknesses, and layup methods are custom-prepared. Debonding defects of different sizes, locations, and shapes are pre-fabricated at the core-panel interface by artificially embedding polytetrafluoroethylene (PTFE) films and release agents. The defective specimens are then installed on specialized equipment such as universal testing machines, and destructive mechanical tests such as static tensile, compression, bending, and shear are conducted sequentially. Test results, including load-displacement curves, stress-strain data, and structural failure modes, are collected using sensors. Finally, the structural ultimate bearing capacity, bending stiffness, shear strength, and other mechanical performance parameters are statistically obtained. The core advantage of this method lies in the reliability and high accuracy of the test data, which can serve as a true value calibration basis for other prediction methods. However, the test specimen preparation process of this scheme is complex, the raw material cost is high, and the economic cost of large-scale working condition testing is extremely high; the controllability of the test conditions is poor, and it is impossible to accurately traverse all parameter combinations of working conditions such as debonding size gradient, debonding position difference, and load amplitude change. Only a small number of discrete working condition tests can be achieved, and the data sample coverage dimension is extremely low; the test cycle is lengthy, and the cycle of single batch of specimen molding, curing, testing, and data processing cannot meet the urgent needs of rapid evaluation and batch screening in engineering sites.

[0006] To overcome the shortcomings of high cost and incomplete coverage of operating conditions in physical testing, the industry generally adopts finite element numerical simulation technology to predict the mechanical properties of core-panel debonding. Relying on computer numerical simulation, a refined geometric model of the sandwich structure consistent with the actual structure is constructed. By defining the layup properties of the composite panel, the constitutive parameters of the core material, and the mechanical parameters of the adhesive interface, cohesive elements and interface contact elements are used to simulate the bonding characteristics and damage evolution of the adhesive interface. After completing model mesh generation, applying load boundary conditions, and setting the solution step size, nonlinear mechanical calculations are performed through an iterative solver to obtain mechanical results such as structural stress distribution, deformation law, buckling critical load, and ultimate failure load. This approach achieves non-destructive, full-condition parametric modeling and analysis, and can partially replace physical testing, effectively expanding the scope of operating conditions. However, this approach has significant technical bottlenecks: high modeling threshold and cumbersome process, requiring professional engineers to complete geometric modeling, mesh convergence verification, interface parameter calibration, and boundary condition optimization, resulting in extremely high labor and time costs; low simulation efficiency, with long time consumption for multi-parameter batch simulations, making rapid prediction impossible.

[0007] Therefore, developing a convenient, efficient, and low-cost intelligent prediction method for the mechanical properties of sandwich structures that can accurately predict different debonding dimensions, core thicknesses, and load conditions has become an urgent technical problem to be solved in this field. Summary of the Invention

[0008] This invention discloses an intelligent prediction method for the debonding mechanical properties of sandwich panel cores. Addressing the technical pain points of existing methods that rely on extensive finite element simulation calculations for predicting the debonding mechanical properties of sandwich panel cores, resulting in high testing costs, long cycles, and difficulty in meeting the rapid evaluation needs of aerospace components, this invention overcomes the shortcomings of traditional methods for testing and simulating the debonding performance of sandwich panel cores, which are characterized by high costs and low efficiency.

[0009] Technical solution: A method for intelligent prediction of the mechanical properties of core debonding in sandwich structure panels, comprising: Step 1: Construct a simulation model library for sandwich structures through finite element parametric modeling; Step 2: Construct a database of mechanical properties for core debonding based on the aforementioned sandwich structure simulation model library; Step 3: Based on the core debonding mechanical properties database, use a deep learning model to intelligently predict the core debonding mechanical properties.

[0010] In summary, this invention discloses an intelligent prediction method for the mechanical properties of core debonding in sandwich panels. First, it utilizes Python scripts to perform batch parametric finite element modeling of composite sandwich panels, setting panel layup parameters, core material geometry, pre-set debonding defect dimensions, and external load parameters as variable parameters to generate a batch finite element model library for sandwich structures. Second, it constructs a simulation database for core debonding failure of sandwich panels based on cohesive elements and the Hashin criterion. Quasi-static tensile / bending simulations of sandwich panels with pre-set debonding defects are conducted using Abaqus to capture key data such as the modulus, strength, and ultimate bearing displacement of the damaged structure. Finally, it builds an improved deep learning model using Python, taking structural geometry, layup, and defect features as model inputs, and the modulus, strength, and buckling load of the damaged structure as model outputs. The model is trained using the simulation dataset, enabling rapid prediction of the mechanical properties of core debonding in sandwich panels under data-driven conditions. This invention avoids the high computational cost of repetitive finite element modeling and calculation, eliminates the need for repeated physical experiments, and can accurately and quickly predict key indicators of core debonding failure in sandwich structures under different defect conditions. It optimizes the early structural design and defect tolerance verification of sandwich structures and is applicable to the engineering design of sandwich composite materials in the fields of aviation, shipbuilding, and rail transportation. Attached Figure Description

[0011] Figure 1 A flowchart of an intelligent prediction method for the mechanical properties of core debonding of a sandwich structure panel provided in this application.

[0012] Figure 2 This is a topological diagram of a deep learning model. Detailed Implementation

[0013] This invention provides an intelligent prediction method for the mechanical properties of core debonding of sandwich structure panels. It combines Abaqus secondary development based on Python, cohesive elements, and two-dimensional Hashin material failure criteria to build a simulation model library in batches, calculate and summarize to obtain a mechanical property database, and rely on deep learning models to achieve rapid and intelligent prediction of the mechanical properties of core debonding of sandwich structure panels without modeling.

[0014] Example 1 like Figure 1 As shown, this application provides an intelligent prediction method for the mechanical properties of core debonding in sandwich structure panels, including: Step 1: Construct a simulation model library for sandwich structures through finite element parametric modeling; The sandwich structure includes composite material upper and lower panels, a lightweight core material, and a pre-formed core panel interface to prevent debonding. The composite material upper and lower panels are made of carbon fiber or glass fiber / epoxy resin composite prepreg; the lightweight core material is any one of aluminum honeycomb, aramid honeycomb, or PVC foam.

[0015] Specifically, step 1 includes: Step 11: Write a Python automated modeling script based on the Abaqus secondary development interface, and set the Python script input variables, including the core structure length L1, core structure width W1, composite material upper and lower panel thickness t1, lightweight core material thickness t2, pre-set core interface debonding defect length L2, and pre-set core interface debonding defect width W2. Step 12: Generate sandwich structure finite element models in batches using Python automated modeling scripts. The sandwich structure finite element models include intact bonding areas and pre-set debonding areas. The intact bonding area uses cohesive units to simulate the bonding characteristics of the core interface, and the pre-set debonding area uses a layered setting with node separation. Specifically, the composite material upper and lower panels of the sandwich structure finite element model are discretized using S4R shell elements, the lightweight core material is discretized using C3D8R solid elements, and the core interface of the intact bonding area is discretized using zero-thickness COH3D8 cohesive elements. Specifically, the simulation loading method of the sandwich structure finite element model is quasi-static three-point bending, with fixed support constraints applied to all nodes at both ends of the length direction of the sandwich structure, and displacement load applied at the mid-span.

[0016] Step 13: Based on the input variables of the Python script, classify and archive all the finite element models of sandwich structures, and build a simulation model library for sandwich structures; In practical applications, the modeling process specifically includes geometric modeling, element generation, material assignment, section definition, interface bonding and defect setting, boundary and load application, etc.

[0017] Step 2: Construct a database of mechanical properties for core debonding based on the aforementioned sandwich structure simulation model library.

[0018] Specifically, step 2 includes: Step 21: Call the Abaqus solver to run the sandwich structure simulation model library in batches; Among them, the cohesive unit is used to determine the debonding initiation and failure of the core interface, and the two-dimensional Hashin material failure criterion is used to calculate the progressive damage failure of the composite panel. Specifically, the cohesive unit utilizes a bilinear traction-separation constitutive relation; the two-dimensional hashin material failure criterion is divided into four failure modes: longitudinal tension, longitudinal compression, transverse tension, and transverse compression. Specifically, during the running of the sandwich structure simulation model library, the cohesive unit is updated iteratively at each step to simulate the entire process of core debonding from initial initiation to debonding failure. Specifically, during the running of the sandwich structure simulation model library, the progressive damage failure of the composite material panel is calculated at each step, simulating the damage evolution process of the sandwich structure panel from stiffness degradation to complete material failure.

[0019] Step 22: Use Python automated post-processing scripts to batch extract the debonding mechanical properties of the core of the sandwich structure simulation model library; Among them, the mechanical properties of the core debonding include characteristic parameters such as maximum displacement D, flexural modulus E and bearing strength S.

[0020] Step 23: According to the Python script input variables, classify and archive the core debonding mechanical properties, and establish a core debonding mechanical property database corresponding to the Python script input variables.

[0021] Step 3: Based on the core debonding mechanical properties database, use a deep learning model to intelligently predict the core debonding mechanical properties.

[0022] Specifically, step 3 includes: Step 31: Construct a deep learning model using Python. The deep learning model consists of an input layer, multiple hidden layers, and an output layer. The input parameters of the input layer are Python script input variables; The number of neurons in multiple hidden layers was determined by trial and error combined with genetic algorithm. The Sigmoid function was selected as the activation function. Gradient descent algorithm was used to optimize the convergence speed of the deep learning model and reduce prediction error. The output parameters of the output layer are the mechanical properties of the core debonding.

[0023] Step 32: Based on the core debonding mechanical properties database, divide the dataset and train a deep learning model; The dataset includes a training set and a test set; Specifically, the dataset is divided as follows: 85% of the data in the core debonding mechanical properties database is used as the training set, and the remaining 15% is used as the test set. The average relative error of the test set is less than 2.5%, which is used as the criterion for model training to pass the convergence test.

[0024] Step 33: Input the structural and defect parameters of the debonding of the core of the new sandwich structure into the deep learning model to intelligently predict the mechanical properties of the core debonding.

[0025] It should be noted that the deep learning model can quickly predict the maximum displacement D, bending modulus E, and bearing strength S of the core debonding of the sandwich structure under new parameters, realizing rapid and intelligent prediction of the mechanical properties of debonding without simulation calculation.

[0026] Example 2 A specific implementation of intelligent prediction of the mechanical properties of core debonding of sandwich structure board: In this embodiment, PVC foam core material is selected, and T700 carbon fiber / epoxy resin composite material is used for the upper and lower layers.

[0027] An automated Python modeling script was written based on the Abaqus secondary development interface, and the input variables of the Python script were set. Among them, the length of the sandwich structure L1∈[100mm, 300mm], the width of the sandwich structure W1∈[50mm, 100mm], the thickness of the upper and lower panels of the composite material t1∈[1mm, 3mm], the thickness of the lightweight core material t2∈[5mm, 10mm], the length of the pre-set core interface debonding defect L2∈[10mm, 30mm], and the width of the pre-set core interface debonding defect W2∈[5mm, 15mm].

[0028] 450 sets of finite element models of sandwich structures were generated in batches using Python automated modeling scripts. The upper and lower composite panels were discretized using S4R shell elements, the lightweight core material using C3D8R solid elements, and the well-bonded areas were simulated using zero-thickness COH3D8 cohesive elements to mimic the bonding characteristics of the core-panel interface. Pre-defined debonding areas were set up with node separation in a layered configuration. A quasi-static three-point bending loading method was used to apply fixed supports to all nodes at both ends of the sandwich structure along its length, while a displacement load was applied at mid-span.

[0029] Based on the input variables from the Python script, all sandwich structure finite element models are categorized and archived to build a sandwich structure simulation model library.

[0030] 450 models were submitted to Abaqus for solving. Cohesive elements were used to determine the initiation and failure of debonding at the core-plate interface. The secondary nominal stress damage initiation criterion and the BK damage propagation criterion were adopted. The interface parameters of the cohesive elements were set as follows: normal elastic modulus 2800 MPa, tangential shear modulus 1014.5 MPa, normal critical tensile strength 5 MPa, tangential critical shear strength 34 MPa, and interface critical strain energy release rate 1 N / mm.

[0031] The failure of the upper and lower panels of the composite material was determined using the two-dimensional Hashin material failure criterion. A linear evolution criterion based on fracture energy was set with the following parameters: fiber tensile strength 1800 MPa, fiber compressive strength 1350 MPa, matrix tensile strength 65 MPa, matrix compressive strength 210 MPa, in-plane shear strength 95 MPa, and fiber tensile fracture energy 75000 J / m. 2 Fiber compressive breaking energy 38000 J / m 2 The tensile fracture energy of the matrix is ​​280 J / m. 2 The matrix compressive fracture energy is 1100 J / m. 2 .

[0032] During the operation of the sandwich structure simulation model library, the cohesive unit is updated iteratively at each step to simulate the entire process of core debonding from initial initiation to debonding failure; the progressive damage failure of the composite material panel is calculated simultaneously to simulate the damage evolution process of the sandwich structure panel from stiffness degradation to complete material failure.

[0033] A Python automated post-processing script was used to extract characteristic parameters such as maximum displacement D, bending modulus E, and bearing strength S from the simulation model library of sandwich structures in batches.

[0034] By inputting variables into a Python script, the mechanical properties of core debonding are classified and archived, forming a database of core debonding mechanical properties containing 450 sets of data.

[0035] A deep learning model is built using Python. The input parameters of the input layer are 6-dimensional Python script input variables: the length L1 of the sandwich structure, the width W1 of the sandwich structure, the thickness t1 of the upper and lower composite panels, the thickness t2 of the lightweight core material, the length L2 of the pre-set debonding defect at the core-panel interface, and the width W2 of the pre-set debonding defect at the core-panel interface. The number of hidden layers, nodes, and transfer functions are adjusted according to the complexity of the problem. The output parameters of the output layer are 3-dimensional core-panel debonding mechanical properties: maximum displacement D, flexural modulus E, and bearing strength S. The topology diagram of the deep learning model is shown below. Figure 2 As shown.

[0036] The deep learning model was trained using 382 sets from the core debonding mechanical properties database and the remaining 68 sets as the test set. The training was stopped when the average relative error of the test set was ≤2.21%. Five sets of new parameters were randomly selected and input into the trained deep learning model for intelligent prediction. The output core debonding mechanical properties were compared with the finite element results. The average prediction error of the three parameters was less than 1.95%, which meets the engineering prediction accuracy.

[0037] In summary, compared with the prior art, the present invention has the following significant advantages: This invention relies on parametric modeling and post-processing with Abaqus and Python to generate hundreds of sets of simulation data in a short time, avoiding repetitive manual modeling and saving a lot of finite element modeling time. This invention employs coupled simulation of cohesive units and the aging criterion of materials to accurately reproduce the process of debonding at the core-panel interface from initiation to failure, as well as the progressive damage process of composite material panels. The database sample data has high accuracy, ensuring the reliability of subsequent deep learning model training. The mature deep learning model trained by this invention can be used independently of finite element software. It can quickly output the debonding mechanical properties of the core plate by only inputting structural and defect parameters, without the need for repeated numerical simulations. This significantly reduces the design cost of defect tolerance for sandwich structures and reduces the amount of physical testing required. The method of this invention is adaptable to a variety of mainstream sandwich core materials such as foam and honeycomb, and is compatible with carbon fiber or glass fiber / epoxy resin composite prepregs. It has strong versatility and is easy to implement in the research and development of aerospace and marine composite products.

[0038] The above detailed embodiments are a description of the present invention. It should not be considered that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the protection scope of the present invention.

Claims

1. A method for intelligently predicting the mechanical properties of core debonding in sandwich structure panels, characterized in that, include: Step 1: Construct a simulation model library for sandwich structures through finite element parametric modeling; Step 2: Construct a database of mechanical properties for core debonding based on the aforementioned sandwich structure simulation model library; Step 3: Based on the core debonding mechanical properties database, use a deep learning model to intelligently predict the core debonding mechanical properties.

2. The method according to claim 1, characterized in that, Step 1 includes: Step 11: Write a Python automated modeling script based on the Abaqus secondary development interface, and set the Python script input variables, including the core structure length L1, core structure width W1, composite material upper and lower panel thickness t1, lightweight core material thickness t2, pre-set core interface debonding defect length L2, and pre-set core interface debonding defect width W2. Step 12: Generate sandwich structure finite element models in batches using Python automated modeling scripts. The sandwich structure finite element models include intact bonding areas and pre-set debonding areas. Step 13: Based on the input variables of the Python script, classify and archive all the finite element models of sandwich structures, and build a simulation model library for sandwich structures.

3. The method according to claim 2, characterized in that, The intact bonding area uses cohesive units to simulate the bonding characteristics of the core interface, and the pre-set debonding area adopts a layered setting with node separation.

4. The method according to claim 2, characterized in that, The composite material upper and lower panels of the sandwich structure finite element model are discretized using S4R shell elements, the lightweight core material is discretized using C3D8R solid elements, and the core interface of the intact bonding area is discretized using zero-thickness COH3D8 cohesive elements.

5. The method according to claim 2, characterized in that, The simulation loading method of the sandwich structure finite element model is quasi-static three-point bending. Fixed support constraints are applied to all nodes at both ends of the length direction of the sandwich structure, and displacement load is applied at the mid-span.

6. The method according to claim 1, characterized in that, Step 2 includes: Step 21: Call the Abaqus solver to run the sandwich structure simulation model library in batches; Step 22: Use Python automated post-processing scripts to batch extract the debonding mechanical properties of the core of the sandwich structure simulation model library; Step 23: According to the Python script input variables, classify and archive the core debonding mechanical properties, and establish a core debonding mechanical property database corresponding to the Python script input variables.

7. The method according to claim 6, characterized in that, Step 2 includes: Cohesive cohesive elements are used to determine the initiation and failure of debonding at the core-panel interface, and the progressive damage failure of the composite panel is calculated using the two-dimensional Hashin material failure criterion; the cohesive cohesive elements utilize bilinear traction-separation constitutive relations. During the operation of the sandwich structure simulation model library, the cohesive unit is updated iteratively at each step to simulate the entire process of core debonding from initial initiation to debonding failure. During the operation of the sandwich structure simulation model library, the progressive damage failure of the composite material panel is calculated at each step, simulating the damage evolution process of the sandwich structure panel from stiffness degradation to complete material failure.

8. The method according to claim 1, characterized in that, Step 3 includes: Step 31: Construct a deep learning model using Python. The deep learning model consists of an input layer, multiple hidden layers, and an output layer. Step 32: Based on the core debonding mechanical properties database, divide the dataset and train a deep learning model; Step 33: Input the structural and defect parameters of the debonding of the core of the new sandwich structure into the deep learning model to intelligently predict the mechanical properties of the core debonding.

9. The method according to claim 8, characterized in that, In step 31, the input parameters of the input layer are Python script input variables; the number of neurons in the multiple hidden layers is determined by trial and error combined with genetic algorithm optimization, the activation function is the Sigmoid function, and the deep learning model convergence speed is optimized by gradient descent algorithm; the output parameters of the output layer are the core debonding mechanical properties.

10. The method according to claim 8, characterized in that, In step 32, The dataset includes a training set and a test set. The dataset is divided as follows: 85% of the data in the core debonding mechanical properties database is used as the training set, and the remaining 15% of the data is used as the test set. The average relative error of the test set is less than 2.5%, which is used as the criterion for model training to pass the convergence test.

Images